Design Doc sign-up via CMS sign up Sunday .... Google (YouTube, Picasa,
Gmail, Docs), Facebook, Flickr. • 100 GB per second is ... Where to? Smart Dust
…
What does the Future Hold? Hakim Weatherspoon CS 3410, Spring 2013 Computer Science Cornell University
Announcements How to improve your grade? Submit a course evaluation and drop lowest homework score • To receive credit, Submit before Tuesday, May 7th
Announcements Final Project Design Doc sign‐up via CMS sign up Sunday, Monday, or Tuesday May 5th, 6th, or 7th Demo Sign‐Up via CMS. sign up Tuesday, May 14th or Wednesday, May 15th CMS submission due: • Due 6:30pm Wednesday, May 15th
Big Picture about the Future
Big Picture How a processor works? How a computer is organized?
register file B
alu
D
memory
D
A
compute jump/branch targets
+4
Instruction Decode
Instruction Fetch IF/ID
ctrl
detect hazard
ID/EX
M
dout
forward unit
Execute
EX/MEM
Memory
ctrl
new pc
din
memory
ctrl
extend
B
control imm
inst
PC
addr
Write‐ Back
MEM/WB
What’s next?
More of Moore
Moore’s Law
Moore’s Law introduced in 1965
• Number of transistors that can be integrated on a single die would double every 18 to 24 months (i.e., grow exponentially with time).
Amazingly visionary 2300 transistors, 1 MHz clock (Intel 4004) ‐ 1971 16 Million transistors (Ultra Sparc III) 42 Million transistors, 2 GHz clock (Intel Xeon) – 2001 55 Million transistors, 3 GHz, 130nm technology, 250mm2 die (Intel Pentium 4) – 2004 • 290+ Million transistors, 3 GHz (Intel Core 2 Duo) – 2007 • 731 Million transistors, 2‐3Ghz (Intel Nehalem) – 2009 • 1.4 Billion transistors, 2‐3Ghz (Intel Ivy Bridge) – 2012 • • • •
Moore’s Law Dual‐core Itanium 2
Ivy Bridge K10
Itanium 2 K8 P4 Atom
486 386 286 8088 8080 4004 8008
Pentium
Why Multicore? Moore’s law • A law about transistors • Smaller means more transistors per die • And smaller means faster too
But: Power consumption growing too…
What to do with all these transistors?
Multi‐core
Multi‐core
http://www.theregister.co.uk/2010/02/03/intel_westmere_ep_preview/
The first transistor
• An Intel Westmere
• on a workbench at AT&T Bell Labs in 1947 • Bardeen, Brattain, and Shockley
– – – – –
1.17 billion transistors 240 square millimeters 32 nanometer: transistor gate width Six processing cores Release date: January 2010
Multi‐core
http://forwardthinking.pcmag.com/none/296972‐intel‐releases‐ivy‐bridge‐first‐processor‐with‐tri‐gate‐transistor
The first transistor
• An Intel Ivy Bridge
• on a workbench at AT&T Bell Labs in 1947 • Bardeen, Brattain, and Shockley
– – – – –
1.4 billion transistors 160 square millimeters 22 nanometer: transistor gate width Up to eight processing cores Release date: April 2012
What to do with all these transistors?
Many‐core and Graphical Processing units
Faster than Moore’s Law One‐pixel polygons (~10M polygons @ 30Hz)
Peak Performance ('s/sec)
Slope ~2.4x/year 10 10 10
9
nVidia G70
(Moore's Law ~ 1.7x/year)
8 UNC/HP PixelFlow SGI IR
7 Division Pxpl6
UNC Pxpl5
10 10
SGI SkyWriter
6 5
SGI VGX
Flat shading
UNC Pxpl4
HP VRX
4
86
HP TVRX Stellar GS1000
SGI GT
HP CRX
10
SGI R‐Monster
ATI Radeon 256
SGI Iris
88
Gouraud shading
90
SGI RE1 E&S F300
Nvidia TNT GeForce E&S 3DLabs SGI Harmony Cobalt Glint Accel/VSIS Voodoo
Megatek E&S Freedom
SGI RE2
Textures
PC Graphics
Division VPX
Antialiasing
92
Year
94
96
Graph courtesy of Professor John Poulton (from Eric Haines)
98
00
AMDs Hybrid CPU/GPU AMD’s Answer: Hybrid CPU/GPU
Cell IBM/Sony/Toshiba Sony Playstation 3 PPE SPEs (synergestic)
Parallelism Must exploit parallelism for performance • Lots of parallelism in graphics applications • Lots of parallelism in scientific computing
SIMD: single instruction, multiple data • Perform same operation in parallel on many data items • Data parallelism
MIMD: multiple instruction, multiple data • Run separate programs in parallel (on different data) • Task parallelism
NVidia Tesla Architecture
Why are GPUs so fast?
FIGURE A.3.1 Direct3D 10 graphics pipeline. Each logical pipeline stage maps to GPU hardware or to a GPU processor. Programmable shader stages are blue, fixed‐function blocks are white, and memory objects are grey. Each stage processes a vertex, geometric primitive, or pixel in a streaming dataflow fashion. Copyright © 2009 Elsevier, Inc. All rights reserved.
Pipelined and parallel Very, very parallel: 128 to 1000 cores
FIGURE A.2.5 Basic unified GPU architecture. Example GPU with 112 streaming processor (SP) cores organized in 14 streaming multiprocessors (SMs); the cores are highly multithreaded. It has the basic Tesla architecture of an NVIDIA GeForce 8800. The processors connect with four 64‐bit‐wide DRAM partitions via an interconnection network. Each SM has eight SP cores, two special function units (SFUs), instruction and constant caches, a multithreaded instruction unit, and a shared memory. Copyright © 2009 Elsevier, Inc. All rights reserved.
General computing with GPUs Can we use these for general computation? Scientific Computing • MATLAB codes
Convex hulls Molecular Dynamics Etc. NVIDIA’s answer: Compute Unified Device Architecture (CUDA) • MATLAB/Fortran/etc. “C for CUDA” GPU Codes
What to do with all these transistors?
Cloud Computing
Cloud Computing Datacenters are becoming a commodity Order online and have it delivered • Datacenter in a box: already set up with commodity hardware & software (Intel, Linux, petabyte of storage) • Plug data, power & cooling and turn on – typically connected via optical fiber – may have network of such datacenters
Cloud Computing = Network of Datacenters
Cloud Computing Enable datacenters to coordinate over vast distances • Optimize availability, disaster tolerance, energy • Without sacrificing performance • “cloud computing”
Drive underlying technological innovations.
Cloud Computing
Vision The promise of the Cloud • A computer utility; a commodity • Catalyst for technology economy • Revolutionizing for health care, financial systems, scientific research, and society
However, cloud platforms today • Entail significant risk: vendor lock‐in vs control • Entail inefficient processes: energy vs performance • Entail poor communication: fiber optics vs COTS endpoint
Example: Energy and Performance Why don’t we save more energy in the cloud? No one deletes data anymore! • Huge amounts of seldom‐accessed data
Data deluge • Google (YouTube, Picasa, Gmail, Docs), Facebook, Flickr • 100 GB per second is faster than hard disk capacity growth! • Max amount of data accessible at one time